Morphological diversity in oleaginous watermelon ( Citrullus mucosospermus ) from the Nangui Abrogoua University germplasm collection

A hundred and seventy-one oleaginous watermelon accessions either collected from different countries or obtained from gene banks were evaluated and compared based on 11 quantitative morphological traits. Principal component analysis on 11 traits revealed 81.19% of the total variability and pointed out variations among accessions, mainly on the basis of fruit size and weight. The dendrogram and factorial discriminant analysis clustered the accessions in four groups. The multivariate analysis of variance showed a significant difference between the four groups and accessions of the group 2 had higher agronomic performances. The confusion matrix gave the details about accessions assignment and pointed out that breeders must have recourse to several multivariate analyses to have better agromorphological classification of accessions. The traits related to fruit size and weight were the most accession distinctive. Consequently, fruit morphological traits could be used for accession identification during collecting missions.

Watermelon (Citrullus vulgaris) (Cucurbitaceae) includes several other economically important species such as cucumber (Cucumis sativus), melon (Cucumis melo L), squash (Cucurbita pepo), calabash (L.siceraria), and pumpkin (Citrullus maxima).C. lanatus is an annual species, which has wild, cultivated, and feral forms.In the finding that the watermelon and its sister species are West African plants suggests that the natural range of watermelon may have extended into Libya or into Egypt during more humid periods of the Pleistocene and Holocene (Schulz, 1987(Schulz, , 1991)).Alternatively, watermelon seeds may have been traded from West Africa to Northern Africa.The illustrations found in Egyptian tombs of watermelon served on a tray suggest that these fruits were eaten raw, perhaps as a dessert.Seeds found in ancient Egyptian tombs, including that of Thutankhamun (Hepper, 1990) should ideally be studied using ancient DNA approaches.
Genetically, the cultivated watermelon is closest to plants from West Africa that represents the gene pool from which watermelon was domesticated (Chomicki and Renner, 2014).The sister species of watermelon is C. mucosospermus, the egusi melon from Nigeria to Senegal and described by Fursa (1983), Che et al. (2003), and Dane and Liu (2007).In consideration of the findings by Fursa (1983)and Chomicki and Renner (2014), the egusitype watermelon is recognized as C. mucosospermus throughout this paper.
The "egusi" watermelon of West Africa is cultivated for its oleaginous and nutritious seeds that are important in the social and cultural live of several peoples (Burkill, 1985;Oyulu, 1977;van der Vossen et al., 2004;Achigan-Dako et al., 2008).For example, dried slightly toasted and ground seeds are used as soup thickener.The "egusi" seeds are reported to be rich in nutrients ~60% lipids and ~30% proteins (Loukou et al., 2007).Edible oil can also be extracted from the seeds.Commonly found in many traditional cropping systems, the plant is well adapted to extremely divergent agro-ecosystems and various cropping systems characterized by minimal inputs.C. lanatus thus represents an excellent plant model for which improved cropping systems implementation can insure the economic prosperity of rural women that are the main producers in tropical Africa.The west African watermelon is generally considered as allogamous, monoecious, entomophilous, and protandrous (Gusmini and Wehner, 2003).This species is named as C. lanatus var.lanatus or as C. lanatus subsp.Mucosospermus (Fursa, 1972) or C. colocynthis (Hutchinson and Dalziel, 1954).
Particularly, in Côte d'Ivoire, indigenous and minor crops for local consumption have almost disappeared in favor of exotic species.Some farmers still continue to produce these crops not far from their home or on small plots cleared after harvest of the main annual crops (yams, corn, peanut, etc.) (Zoro Bi et al., 2003).Thus, some of them such as oleaginous cucurbits are still found in the market.According to Zoro Bi et al. (2003) and Djè et al. (2006), C. mucosospermus (but identified as C. lanatus by these authors in Côte d'Ivoire) is the species most widely distributed and most commercialized in Côte d'Ivoire.This plant is mainly grown in savanna areas in the North, in the pre-forest areas in Central, East, and South (Djè et al., 2006).Two distinct cultigroups have been described on the basis of the morphology of seeds.The first one (wlèwlè) is characterized by glossy seeds with a tapered proximal extremity.This morphotype is subdivided into three cultivars according to the size of the seeds (Djè et al., 2006).Fruit color of this cultigroup, is almost uniform, greenish or whitish.The second (bebu), with one cultivar has green fruit also, but with dark longitudinal bands.The seeds are heavier and have a flat ovoid shape with rugged and thick ends (Zoro Bi et al., 2003).Although the oleaginous type of Citrullus is an economical important crop in west Africa.Few studies were conducted on genetic resources of this plant (Ndabalishye, 1995;Djè et al., 2006) previously reported to be morphological differentiated especially in terms of fruit and seed characters.
The objectives of the present study were to survey and analyze variations in plant, fruit and seed morphology in C. lanatus oleaginous type collection from Côte d'Ivoire.This represents the first study on collection and characterization of C. lanatus oleaginous type germplasm in Nangui Abrogoua University.

Plant and collection sites
In this study, 171 accessions of C. lanatus oleaginous type were used.They were provided from seven countries: Benin, Côte d'Ivoire, France, Ghana, Nigeria, Togo, and Turkey (Table 1).The samples were collected from farmers' stock, genebank or purchased from local seed markets.The selected accessions were divided into four cultivars defined on the basis of seed size and designated small-, medium-, big-seeded, and thickened margin seeds.Thickened margin seeds size varies between 153 and 205 mm², big seeds from 125 to 151 mm², medium seeds from 86 to 110 mm², and small seeds from 41 to 52 mm² (Djè et al., 2006).Accessions (2 to 101) were sampled per cultivar, according to seeds availability (Table 2).

Study site
The trials were regularly monitored throughout the growing season in 2011 and 2012 (May to November).Farm experiment was conducted in the village of Manfla, located in the centre (6°49′34.38″N,5°43′47.68″W)400 km North Abidjan (Côte d'Ivoire).
Annual rainfall varies from 800 to 1400 mm with a long-term mean of 1200 mm, and the annual mean temperature is 27°C.A complete description of study site was done by Kouassi and Zoro Bi (2010).

Experimental design
Planting was done according to a completely randomized block design, with seven replications.Each replicate consisted of a 24×27 m containing 70 holes at a depth of 3 cm, resulting in 7 holes per accession.The planting distance was 3 m between and within rows with 1.5 m of edges.A manual weeding was carried out during plant development to prevent weed invasion.Disease and pest control was carried out using a carbamate-based insecticide applied when necessary.

Traits measurement
Accessions were examined using 11 quantitative traits subdivided into four vegetative and seven yield characters selected from standard descriptors for cucurbits (Maggs-Kölling et al., 2000;Marr et al., 2007;Morimoto et al., 2005;Koffi et al., 2008).Data were measured in five plants and five fruits per accession.Vegetative traits measured were limb length (LL), limb width (LWI), plant length (PL), and number of branches from the central taproot (BN) for each plant.Yield characters measured were weight (FWE), length (FL), width (FWI), seeds cavity diameter (SCD), seeds number (SN), and seeds weight (SW).Seed traits analyzed included 100-seeds weight (100-SWE).Measurements on 100seeds weight were scored using five individuals randomly selected on each plant (Table 3).

Data analysis
Statistical analyses were carried out with STATISTICA software   , 2005).Principal Component Analysis (PCA) was applied to identify most discriminant parameters followed by hierarchical cluster analysis using unweighted pairgroup method using arithmetic average (UPGMA) method.Multivariate analysis of variance (MANOVA) and a factorial discriminant analysis (FDA) were performed to check difference between the variable means for each group obtained with the clustering analysis.A confusion matrix was constructed, to check the reliability of groups as defined by the hierarchical clustering.

Descriptors variation
The relative discriminating capacity of the PCA is shown by their Eigenvalues.The result of the PCA (

Accessions structuration
The unweighted pair group method arithmetic (UPGMA) using a Euclidian distances matrix, subdivided the accessions analyzed into four major groups.These groups differ in the distance by 30 (Figure 1).All groups are mixture of accession of various origin, cultivar or cultigroup.
The highest number of accessions was located in cluster III with 53 accessions, followed by clusters II and I with 48 and 46 accessions, respectively.Cluster IV had 24 accessions.Comparison of groups using the MANOVA showed significant differences between the four groups (F= 19.91; P<0.001).This is due to all the 11 characters measured.The ANOVA showed significant difference among groups (Table 5).All parameters revealed a partial distinction of groups.
Discriminant functions resulting from variables to classify the units formed on the four groups are shown in Table 6.The two first canonical variables which had an Eigenvalue higher than 1 were retained to describe the total variability of the accessions.The two factors selected together, explained 87.54% of the total variation.The first canonical variable, FDA1, described 58.33% of the total variation.This component was positively correlated with the number of branches and was negatively correlated with limb width, number of seed per fruit, and 100-seeds weight.The second canonical variable, FDA2, explained 29.21% of the total variation and was mostly and positively constructed from fruit diameter.Figure 2 shows the position of the accessions in relation with the first two discriminant factors.Based on these analyses, accessions of group 2 were noted to have the biggest leaves.The smallest leaves were observed in accession of group 1.The most number of 0 10 20 30 40 50 60  the longest plants were observed in group 4. Groups 1 and 2 have the larger and heavier fruit.The best seed yield is obtained with accessions of group 2, but 100 seed-weight was higher in groups 1 and 3.According to confusion matrix, the composition of groups as defined by the hierarchical clustering is different.Group 1 contains 41 accessions instead of 46.Accessions CL-146 and CL-172 have been reclassified in group 2. Similarly, accessions CL-142 and CL-085 have been reclassified in group 3 and accession CL-131 has been reclassified in group 4. Group 2 consists of 44 instead of 48 accessions.An enrollment of 50 accessions belongs to group 3 instead of 53.Group 4 includes 23 accessions instead of 24 (Table 7).

DISCUSSION
To maintain, evaluate and utilize germplasm efficiently, it is important to investigate the extent of genetic diversity available.Morphological characterization is an important step in the description and classification of crop germplasm, because a breeding program mainly depends upon the magnitude of morpho-phenological variability (Smith    , 1991).
In this study, 171 accessions of Citrullus of Nangui University collection were characterized based on morphological traits.In the analysis made to estimate the relative contributions of the different traits studied towards the overall phenotypic variation among the 171 accessions, a total of five principal components (PCs), explained 81.2% of variability.Fruit-related parameters were the most discriminant.In fact, fruit length, fruit diameter, and seed cavity diameter were parameters which contributed more to the first principal component (24%), implying that 24% of the variability was explained by these parameters.These results suggest that fruit size is a useful criterion for separating Citrullus accessions.In the cucurbit family, the significant contribution of the fruit traits to morphological variability has been reported for watermelon (Gusmini and Wehner, 2003;Maggs-Kölling et al., 2000).According to Achigan- Dako (2015) diagnostic morphological markers among species include fruit such as fruit size.
The least discriminant parameters were numbers of seeds and seeds weight.This finding is due to the fact that medium-sized seeds represented 60% of sample analyzed.Contrary to these results, Achigan- Dako et al. (2015) demonstrated that other major differentiation traits is related to seed.
Although our accessions originated from seven different countries, they did not cluster according to geographical distribution.This may be suggestive of the single ancestry of the analyzed plants (Dane and Liu, 2007).This may be because genetically, the cultivated watermelon is closest to plants from West Africa that represents the gene pool from which watermelon was domesticated (Chomicki and Renner, 2014).The sister species of watermelon is C. mucosospermus, the egusi melon, ranging from Nigeria to Senegal (Renner et al., 2014).Plants from Benin closely resemble the collection type of the name C. mucosospermus (which is from Ghana).In order to explain the low genetic diversity of C. mucosospermus (but identified as C. lanatus), Dane and Liu (2007) demonstrated that the watermelon varieties have diverged into small populations.Moreover, the domestication and the resulting selection may have Gbotto et al. 927 reduced the genetic diversity from the original watermelon crops.Levi et al. (2001) also revealed that cultivated watermelon exhibited narrow genetic base as a results of many years cultivation and selection for specific qualities.Morphological and genetic characterization of Hungarian and Turkish accessions of watermelon (including C. lanatus and C. colocynthis) revealed that germplasm resources present a wide range of diversity (Sari et al., 2005;Solmaz and Sari, 2009;Solmaz et al., 2010).However, accessions of the two countries show many similarities and therefore cannot be separated clearly (Szamosi et al., 2009).In the present study, comparison of the four groups using a MANOVA combined with factorial discriminant analysis, allowed the classification of accessions according to agromorphological characters.Accessions of group 2 had the biggest leaves.The smallest leaves were observed in accession of group 1.The largest number of ramification and longest plants were observed in group 4.
Groups 1 and 2 have the largest and heaviest fruits.The best seed yield is obtained with accessions of group 2, but 100 seed-weight was higher in groups 1 and 3 than in the two other groups.Morphological variation analysis in landraces of C. mucosospermus (but identified as C. lanatus by authors in Côte d'Ivoire) also allowed to classify accessions based on fruit and seed traits (Adjoumani et al., 2012).
Although, the Hierarchical Cluster Analysis gave a clear separation of accessions in distinct groups, the confusion matrix displayed a reclassification of some accessions.Accessions CL-146 and CL-172 has been reclassified in group 2. Accessions CL-142 and CL-085 has been reclassified in group 3 and accession CL-131 has been reclassified in group 4.
Traditionally, watermelon breeders focus on fruit quality traits.These traits include sugar content, flesh color, fruit size, and rind patterns (Hashizume et al., 2003).Investigations of the inheritance of fruit morphology and quality traits date back as far at the 1930's (Porter, 1933(Porter, , 1937)).Since then, many efforts have been made to better understand traits associated with watermelon fruit quality and morphology.Many genes have been described that control the internal fruit quality and morphology in watermelon (Guner and Wehner, 2004).An internal fruit characteristic that has received attention is the Brix value.
A linkage map was developed and a QTL that accounts for 19% of the variation in Brix was mapped on linkage group (LG) 8 (Prothro, 2010).Fruit shape and weight are important external characteristics that breeders must consider when developing watermelon cultivars.
Watermelon shape can be classified as being either round, oval, blocky or elongate (Wehner et al., 2001).The only gene described that controls fruit shape is the elongate fruit gene (O).Fruit weight has recently become an important consideration for breeders due to increased seeds taken for a given individual after drying 100-seeds weight per fruit (g) (100-SWE) Weight of 100 seeds taken for a given individual after drying A measurement on five organs per plant; b measurement on each plant per accession.package version 7.1(StatSoft

Table 2 .
Origins and size of Citrullus mucosospermus samples used for morphological traits analysis.

Table 3 .
List of descriptors used for characterization of Citrullus lanatus samples used for morphological traits analysis.

Table 4
number of branches, and plant length for PC3, number of seeds per fruit and 100-seeds weight per fruit (100-SWE) for PC4 and seeds weight per fruit for PC5.These results indicated that PC3 was determined by individuals of large size plant.PC4 and PC5 were determined by most productive individuals.

Table 4 .
Correlation coefficients explained by the first five principal components (PCs) for 11 traits analysed in 171 accessions of Citrullus mucosospermus.

Table 5 .
Agro-morphological characteristic of 11 traits analyzed in the four groups provided by the dendrogram with UPGMA methods.

Table 6 .
Eigenvectors, eigenvalues, and inertia percentage explained by the two first canonical variables for 11 traits analysed in 171 accessions of Citrullus mucosospermus.

Table 7 .
Classification matrix based on the groups obtained by agromorphological characters.